pH-Dependence of Protein Stability: Absolute Electrostatic Free

pH-Dependence of Protein Stability: Absolute Electrostatic Free Energy Differences between Conformations. Michael Schaefer, Michael .... On the Use of...
0 downloads 4 Views 383KB Size
J. Phys. Chem. B 1997, 101, 1663-1683

1663

pH-Dependence of Protein Stability: Absolute Electrostatic Free Energy Differences between Conformations† Michael Schaefer, Michael Sommer, and Martin Karplus* Department of Chemistry, HarVard UniVersity, 12 Oxford Street, Cambridge, Massachusetts 02138, and Laboratoire de Chimie Biophysique, Institut le Bel, UniVersite´ Louis Pasteur, 4 rue Blaise Pascal, 67000 Strasbourg, France ReceiVed: September 26, 1996; In Final Form: December 5, 1996X

A method for calculating the absolute electrostatic free energy of a titrating system as a function of pH is proposed, and a concise formula for the free energy is presented. Based on the theory of linked functions, the electrostatic free energy is calculated by integration of the titration curve. The approach uses pH ) ∞, at which the system is in the unprotonated state, as the reference pH for the integration. The finite-difference Poisson-Boltzmann method is used for the electrostatic free energy, and the titration curve is obtained by the Monte Carlo approach of Beroza et al.1 The method is applied to the native and denatured state of hen egg-white lysozyme in aqueous solution. A dielectric constant of 20 is assigned to the protein interior, in accord with the work of Antosiewicz et al.2 X-ray structures are used for the native state, and an extended β-structure is used for the unfolded reference state; good agreement with experiment is obtained. Comparison of the results for the extended β-structure and for a “null” model of noninteracting sites for the unfolded state shows significant differences. This indicates that there are important interactions between titrating sites in the unfolded state, in agreement with recent experimental estimates by Oliveberg et al.3 A number of structures obtained by in Vacuo minimization of the native and unfolded protein are compared, and it is shown that the absolute pH-stability curve has a strong conformational dependence, although the relative stability curve does not. Calculations of absolute free energy differences, rather than relative changes as a function of pH, are of general interest; for example, they can be used to study the binding energy of ligands involving small titratable compounds.

1. Introduction The pKa’s and pH-dependent ionization states of titratable groups play an essential role in the catalytic activity4 and stability of proteins.5-8 Approaches for calculating pKa’s, protein titration curves, and pH-dependent protein stability have, therefore, been the subject of theoretical studies for many decades.9-12 All theoretical approaches have in common the assumption that the observed shift of the pKa of a titrating amino-acid side chain in a protein, relative to the pKa of the isolated amino acid, is dominated by electrostatic interactions, namely, the interactions with other titrating sites, with the surrounding protein residues, and with the aqueous environment including salts. This hypothesis is supported by the experimental observations of pKa-shifts due to point mutations13,14 and due to changes in the ionic strength of the solution,14,15 as well as by the fact that the titration curve of an unfolded protein is approximately equal to the sum of the titration curves of the constituent sites.16,17 For the calculation of the pKa’s, titration curve, and pHdependent stability of a protein, one or more structures must be selected to represent the different states. For determining the pH-dependence of the stability, which is of primary interest in this paper, the crystallographically determined native structure and an appropriate model of the unfolded protein are required. Some approaches have avoided the requirement for an unfolded structure by restricting themselves to the calculation of relative stabilities.2,18 By “relative stability”, we mean that they have assumed that the unfolded state of a protein can be treated as a † This work was partly supported by the National Institutes of Health. M.Sch. is currently supported by a research training grant within the Biotechnology Programme of the European Community. X Abstract published in AdVance ACS Abstracts, January 15, 1997.

S1089-5647(96)02972-0 CCC: $14.00

state where all sites of the protein possess their standard pKa’s and titrate independently (“zero interaction model”18 or “null model”2). With this assumption, no unfolded reference structure with its associated Coulomb and electrostatic solvation energy is required. The stability is then calculated relative to an arbitrary pH reference value, usually pH ) 0, where the free energy difference ∆∆G is set to zero. The main advantage of the null model is that the titration curve and the free energy can be calculated from simple analytic formulas. Another reason for the use of the null model in previous work is that the geometric dimension of unfolded protein models, e.g., an extended structure, leads to accuracy problems in the calculation of electrostatic energies with the finite-difference PoissonBoltzmann method (see below). The stability calculation described in this paper is concerned with the effect of pH on the absolute stability. It can be divided in three steps: (1) for each structure, the relevant electrostatic energies must be calculated (“electrostatic energy calculation”); (2) the pH-dependent average charge states of the titrating sites must be determined (“titration calculation”); and (3) the results of the titration calculations for the different conformers are used to obtain the pH-dependent free energy difference (“stability calculation”). Significant theoretical and algorithmic improvements in the methods for calculating the electrostatic energy2,12,19-21 and titration curves1,21-23 have been reported in recent years. They are based on the finite-difference Poisson-Boltzmann (FDPB) method24,25 and the finite-element method26,27 for the calculation of electrostatic energies. The FDPB method is used in this work. It allows us to include the effects of the solvent and ionic strength on the self-energy and interaction energy of the charges, and it has been employed successfully in numerous © 1997 American Chemical Society

1664 J. Phys. Chem. B, Vol. 101, No. 9, 1997 studies of proteins.28-31 A comparison between different methods for calculating electrostatic energies in the framework of pKa calculations will be presented elsewhere.32 In an exact treatment of the titration curve, which involves the calculation of the generating (partition) function for all possible charge states of the system, the computing time increases exponentially with the number of titratable sites.1,22 Using the fastest computers that are available today, the exact treatment is limited to systems with up to 25-30 titrating sites. Various approximation methods for the calculation of titration curves for large systems have, therefore, been proposed, e.g., the reduced site (RS) approximation,22 a Monte Carlo (MC) titration method,1 and a cluster method.23 The major disadvantage of the RS approximation is that it is limited to systems where less than 25-30 sites (i.e., the number of sites that can be treated exactly) have similar pKa’s. The cluster method has been shown to yield titration curves at low computational expense (several seconds on a Silicon Graphics Iris 4D/VGX for systems with up to 123 sites23); however, its accuracy in comparison with other titration methods has not been fully tested. Consequently, we use the MC titration method, which has the advantage of providing well-defined error bounds, while the required computation time, which depends quadratically on the number of sites, is still small when compared with the time that is necessary to perform the electrostatic energy calculations with the FDPB method.1,23 The calculation of a protein pH-stability curve based on the titration curves of the native and unfolded states is possible because, at a given pH, the derivative of the free energy difference between the two states is proportional to the difference between the average charges of the conformers, ∂(∆G)/ ∂(pH) ∝ ∆ 〈Q〉. This simple relation between the free energy and the titration curve of a protein results from the theory of linked functions.33-35 It follows that the stability curve of a protein can be obtained by integrating the difference between the titration curves of the native and unfolded states,2,11,18 an approach that is termed the “titration curve integration” method in the remainder of this paper. The method for calculating the absolute, pH-dependent stability of proteins uses an extended structure of the protein as the unfolded reference state. The requirement of defining an arbitrary reference pH and free energy difference in the titration curve integration method is eliminated by integrating from pH ) +∞, where the partition function of both the native and the unfolded structure corresponds to the fully unprotonated state. The integration from pH ) -∞ where the protonated state dominates would also be possible and should lead to identical results, provided that the electrostatic free energies of the protonated and unprotonated states are calculated with sufficient accuracy. The integration from pH ) ∞ is justified because the calculated free energy is independent of the integration path over which it is determined, given a welldefined reference state. In this respect, the fact that a native protein structure is not stable under very low or high pH conditions is irrelevant. Contributions, other than the electrostatic term, to the free energy difference between the conformational states of a molecule are significant. They include entropic contributions, van der Waals energies, and the hydrophobic effect.18,36 They are not considered here because this work is aimed exclusively at demonstrating that it is possible to calculate the pH-dependent electrostatic contribution to the absolute free energy. Further, as already mentioned, the pH-dependence is expected to be dominated by the electrostatic contribution. Although the theory presented in section 2 provides a basis for considerations of

Schaefer et al. pH-dependent protein stability that include all energetic and entropic contributions, such an analysis is beyond the scope of this paper. Also, a methodology for calculating pKa’s that accounts for conformational changes of a protein in response to changes in the ionization state would be more correct than the calculation of site-site interaction energies for a single conformer, as pointed out by Bashford and Karplus.12 Approximations to take account of multiple conformers have been introduced.18,37 However, their accuracy is difficult to assess because no Boltzmann weighting of the different conformers associated with a given structure was made. For numerical calculations, it is necessary to choose a dielectric constant for the protein interior. Although various choices have been made,2,12,23 we follow Antosiewicz et al.,2 who have found that an electrostatic model where the protein interior is assigned a dielectric constant of i ) 20 yields good results; they suggest that such a large dielectric constant accounts implicitly for the conformational flexibility of a protein; that is, in conjunction with the use of a single conformer to represent the native state of a protein, this approach has been shown to yield good agreement between calculated and experimental pKa’s for a number of proteins.2 It is also possible that its success is based on the fact that most of the titrating groups in the proteins that were studied are solvated. The importance of conformational equilibria to the calculation of pKa’s and the accuracy of approaches that account for conformational change implicitly and explictly will be addressed in a separate study.38 It should be pointed out that the theory presented in this paper is valid at thermodynamic equilibrium. Use of continuum electrostatics to take account of the solvent effect in molecular dynamics calculations, for example, would require consideration of the time scale of the protonation and deprotonation reaction, relative to the motion of interest.39 In section 2, we develop the theory for calculating pKa’s, titration curves, and stability curves based on the assumption that all sites have only two protonation states, protonated and unprotonated (“two-state model”). The pH and protonation state-dependent free energy of a titrating system is expressed in a form that is more concise than formulas available from the literature; some details are given in Appendix A. It allows straightforward extension of the theory to include conformational averages and more complex titrating sites. In Appendix B, formulas for the general case of multiple (>2) protonation states of the sites are given as a basis for future work.40 In section 3, we describe the methods used for preparing the native and unfolded input structures, for the calculation of electrostatic energies with the FDPB program UHBD,41 and for the titration calculation with a Monte Carlo titration program provided by Beroza et al.1 In section 4, we apply the method for calculating absolute pH-dependent free energy differences to the protein lysozyme. Lysozyme is used because experimental data are available and it has been the object of many calculations of electrostatic properties. It is shown that the agreement between the calculated and experimental pKa’s of lysozyme is best when a dielectric constant of i ) 20 is used for the protein interior, both for the crystallographic structure and for minimized structures of the protein. We present the calculated relative stability curve of lysozyme, using the null model and several explicit models of the unfolded state. For the explicit unfolded models, absolute stability curves are also calculated and compared with the relative stability curves. A strong dependence of the absolute stability, as compared to the relative

pH-Dependence of Protein Stability stability, on the degree of minimization of the native and unfolded structures is demonstrated. We discuss extensions of the present approach in section 5. Possible solutions for the problem of conformational equilibria with respect to pKa’s and stability curves are considered. The accuracy and self-consistency of the methodology related to the use of the FDPB method for calculating electrostatic energies are discussed, and the applicability of the present approach to calculating the absolute, pH-dependent free energy of binding between molecules with titrating groups is outlined. In Appendix C, the problem of inconsistencies in the present treatment of nonbonded Coulomb interactions, which is common to all published work on protein pKa’s employing the FDPB method,2,12,21 is discussed in detail. 2. Theory 2.1. Definitions. We define the standard pKa of a titrating side chain in a protein as the experimental pKa of the amino acid (“X”) as part of a small peptide (the standard peptide) with electrically neutral atom groups except for the site itself, e.g., the peptide Ala-X-Ala with N- and C-terminal blocking groups.4,42,43 Correspondingly, the standard pKa’s of the Nterminal ammonium group and of the C-terminal carboxyl group are the experimental pKa’s of these groups in an otherwise electrically neutral peptide. The standard pKa is denoted by in the following. The observed pKa of a titrating atom pKstnd a group in a protein, termed the “effective pKa”, may differ from the standard pKa by several pK-units. For example, the largest measured pK-shift for titrating sites in the proteins lysozyme,12,44,45 ribonuclease A,46 and myoglobin47 is about 2.5 pK-units. A third pKa that is commonly used in theoretical studies is the “intrinsic pKa” of a site, pKintr a , which is the hypothetical pKa of the site assuming that all other titrating sites in the system are fixed in their electrically neutral state. In a system with only one titrating group, the effective pKa of the group is equal to the pKintr a . For calculating the pH-dependent properties of a system, a choice of the titrating sites that are included in the analysis has to be made. In this choice, it is possible to exclude atom groups whose standard pKa’s are far from the pH-range of interest, e.g., ) 13.6 in considerations of the Ser hydroxyl group with pKstnd a protein stability around pH ) 7. Each titrating site is composed of a set of atoms whose partial charges depend on the protonation state of the site, e.g., all sidechain atoms in the case of aspartate. The atoms of the system that are not part of any site are termed the “background atoms” or “background charges”.12 By definition, the background atoms are assumed to have pH-independent partial charges, even though the theory that is developed in this work makes use of infinite pH-limits; for example, if peptide N-H groups are excluded from the set of titrating sites, it is assumed that they are not ionized in the limit pH f ∞. This does not cause any difficulties and it is understood that the pH-dependent properties of a system are defined relative to a chosen set of titrating sites. 2.2. Electrostatic Free Energy of a Protonation State. Given a protein with N titratable sites, we describe the protonation state by a vector js with N components s1, ..., sN. To derive an equation for the pH-dependent energy of protonation state js, we consider the case where all sites have only the two ionization states “unprotonated” and “protonated”, termed the “two-state model” in the following. In Appendix B, the formulas in this section are generalized to be applicable to the case of multiple states. If the components si are defined

J. Phys. Chem. B, Vol. 101, No. 9, 1997 1665 according to

si )

{

0, if site i is unprotonated 1, if site i is protonated

(1)

the pH-dependent free energy difference between the protein in protonation state js and the fully unprotonated state is12 N

∆G(sj,pH) ) (ln 10)kBT

si(pH - pKintr ∑ a,i ) + i)1 N

[qi(si) qj(sj) - qi(0) qj(0)] Wij ∑ i 16) and by -12 to -13.5 kcal/mol relative to the low pH-limit (pH < 1.5). This result, which was obtained using the PDB structure for the native protein and the extended model for the unfolded state, is in reasonable agreement with the experimental observation that the stability of lysozyme varies by -9.2 kcal/mol when the pH is increased from 1.5 to 7. The result obtained with the extended model for the unfolded state is better than the null model of noninteracting sites, which leads to a predicted change in the stability by -16.5 kcal/mol when the pH is varied from 1.5 to 7. There are significant interactions between the titrating residues for the fully extended structure of lysozyme, which represents a limiting case of the denatured protein. In the case of lysozyme, the interactions in the extended structure give rise to a net stabilization of -1.5 to -4.7 kcal/mol at pH ) 7, relative to the null model. By comparing the calculated pH-stability of various models for the native and unfolded structure of lysozyme, we have demonstrated that there is a strong dependence of the absolute pH-stability on the choice of structures, particularly that of the denatured state. There is a variation of as much as 25 kcal/ mol in ∆GAB(0h), the value for the fully unprotonated state. However, other than this “offset” of the absolute stability curve at pH ) ∞, the pH-stability of lysozyme is almost unaffected by the variation in the structures used for the calculations; that is, the absolute stability curves were found to be parallel over the entire pH range. The data in Table 5 show that the variation of the folding free energy ∆EAB jref) for the electrically neutral el (s ref 0 charge state (sj ) js ) as a function of the native and unfolded structures A and B is the same as the variation of the folding free energy for the standard charge state. Since only the polar atom groups, in particular hydrogen bonds, contribute to the electrostatic free energy of the neutral charge state, it follows that small structural changes have a strong effect on the interactions of polar groups (and, consequently, on the electrostatic free energy of the unprotonated state), whereas the pKa’s and the titration curve of a structure (and, consequently, the relative pH-dependent free energy difference between conformers) are less sensitive to conformational change. The method for calculating pKa’s and absolute pH-stability curves introduced in this paper can be improved with respect to several aspects: First, a continuum electrostatic model where the protein “core” (core ≈ 2-4) and atom groups at the surface (surf ≈ 20-80) are treated differently can be introduced;66 that is, the assignment of more appropriate dielectric constants may lead to improved predictions, particularly for the pKa’s of buried sites.47 Second, standard electrostatic free energies of protonating titrating sites in the model peptides could be used, instead of the current practice of using the energies calculated for individual model compounds with different structures for each site. This would eliminate the error due to the inconsistent treatment of intraresidue Coulomb interactions in the present methodology (see Appendix C). Third, a more detailed treatment of the interaction between titrating sites (including the background atoms) would result from the consideration of multiple protonation states, instead of the average-charge twostate model employed here. Fourth, conformational averages could be introduced for the calculation of the pH-dependent electrostatic free energy of structures, particularly for multiple side-chain conformers.12,37 This would reduce the strong dependence of the absolute pH-stability on the choice of structures observed in this study. The average over multiple

Schaefer et al. conformations could be made at various levels of the theory, e.g., at the conformational level, at the electrostatic free energy level, at the generating function level, and at the titration curve level. The various improvements in the methodology described here are under investigation.78 Even though this paper has been concerned with the pHstability of proteins, the present theory can be directly applied to the calculation of the absolute, pH-dependent binding free energy of a complex of two molecules with titrating sites, e.g., the binding energy for the inhibitor of a protein, given the structure of the complex. For the calculation of binding energies, the bound/unbound states correspond to the folded/ unfolded state (i.e., conformer A/B) in the present work. Since in the unbound state the two molecules of a complex titrate independently, the titration curve for the unbound state is given by the sum of the titration curves of the constituent molecules. Correspondingly, the electrostatic free energy of the unbound state is the sum of the electrostatic free energies (Coulomb energy plus solvation free energy, see Figure 9) of the two molecules. On the basis of the present work, it is thus possible to predict the total electrostatic contribution to the binding free energy of a complex of two molecules at any given pH, including the contribution from the interaction of polar atom groups. To obtain the full binding interaction, other contributions (e.g., hydrophobic and van der Waals terms, configurational entropy) have to be included. Acknowledgment. We are grateful to J. A. McCammon and co-workers for making available a copy of the UHBD41 program, and to P. Beroza et al.1 for a MC titration program. We thank H. van Vlijmen for many helpful discussions. Appendix A. Energy of a Protonation State To derive eq 8 for the electrostatic free energy of a protonation state, ∆G(sj,pH), we first rewrite eq 5 for the pKstnd shift between the pKintr a,i and the pKa,i of site i (see Figure 1) in terms of atom group contributions, stnd -(ln 10)kBT(pKintr a,i - pKa,i )

) ∆E1 - ∆E0 i ) [Eel(si)1,sj*i)s0j ) - EM el (si)1)] i [Eel(si)0,sj*i)s0j ) - EM el (si)0)]

) 1/2[∆Eii(1,1) - ∆Eii(0,0)] + ∆Ei0(1,0) - ∆Ei0(0,0) + N

[Eij(1,s0j ) - Eij(0,s0j )] ∑ j*i

(A-1)

where s0j is defined as the uncharged state of site j and where i ∆Eii(si,si) ) Eii(si,si) - EM ii (si,si) i ∆Ei0(si,0) ) Ei0(si,0) - EM i0 (si,0)

(A-2)

The sum over j * i is over all titrating sites except i and does not include the interaction with the background atoms (site index 0), which is accounted for explictly (∆Ei0 terms). In going from eq 5 to eq A-1, the self-energy of the background atoms, (1/2)E00, in the protein and in the model compound cancels because it contributes equally to the energy of the protein (model compound) with site i protonated or unprotonated. From eq 1 and the definition of the net charge of site i in charge state si and in the unprotonated state, qi(si) and qi(0), it follows that qi(si) ) qi(0) + si. Furthermore, the uncharged

pH-Dependence of Protein Stability

J. Phys. Chem. B, Vol. 101, No. 9, 1997 1681

state s0i and the net charge qi(0) of site i in the unprotonated state are related according to qi(0), ) - s0i , since in the twostate model

qi(0) )

{

- 1, 0,

if site i anionic if site i cationic 1, s0i ) 0,

{

if site i anionic (A-3) if site i cationic

Using eq A-1 for pKintr a,i and eq 3 for Wij, it is thus possible to rewrite eq 2 for the energy of a protonation state in the form N

∆G(sj,pH) ) (ln 10)kBT

si(pH - pKstnd ∑ a,i ) + i)1

N

si[1/2(∆Eii(1,1) - ∆Eii(0,0)) + ∆Ei0(1,0) ∑ i)1 N

∆Ei0(0,0)] +

{si[Eij(1,s0j ) - Eij(0,s0j )] + sj[Eij(s0i ,1) ∑ i